# Video Translation: Subtitles, Dubbing or AI Lip Sync (2026) > Video translation explained: the three levels (subtitles, dubbing, AI lip sync), the step-by-step method, a worked example and the mistakes to avoid. _Source: https://professionalstoolkit.com/articles/video-translation — The Professional's Toolkit · updated November 2026_ _Part of: [Translate a video into another language, with subtitles or a dubbed voice](https://professionalstoolkit.com/solutions/translate-a-video) — the guide, the do-it-yourself version and the tools that do it_ --- Video translation means carrying a video's spoken words, and often its on-screen text, into another language so a new audience can follow it. It comes in three levels: translated subtitles, a dubbed voice, or a dubbed voice with AI lip sync that reshapes the speaker's mouth to match. The method never changes. Get an accurate transcript, translate it, have a fluent speaker review it, then pick the output your viewers need. Subtitles are the fastest way to translate video. Dubbing and lip sync add reach, and they add risk. **20 CPS** — Netflix's reading-speed limit for adult subtitles: up to 20 characters per second - **42** — Characters per subtitle line (Netflix) - **120 min** — Longest video YouTube auto-dubs - **20-40%** — How much longer translated text can run (Synthesia) ## Subtitles, dubbing or lip sync: the three levels of video translation Each level builds on the one before it, so the choice is less about technology and more about how people will watch. A clip scrolled on mute needs text. A training course watched with headphones needs a voice. A presenter talking straight to camera may need both a voice and a mouth that matches it. ### Translated subtitles Subtitles keep the original voice and add text in the viewer's language. They are the lightest level: a caption file, usually SRT or VTT, sits next to the video and the player switches languages on demand. The catch is reading speed. Netflix's English style guide caps a subtitle line at 42 characters and adult reading speed at up to 20 characters per second [1], so a good translation is often condensed rather than copied word for word. ### A dubbed voice Dubbing replaces the spoken audio with speech in the target language. AI dubbing does it in one pass: it transcribes, translates, then speaks the new script with a synthetic voice or a clone of the original speaker. Viewers listen instead of read, which matters for training, for younger audiences and for anyone watching on a small screen. **Timing is the hard part.** Translated speech rarely matches the original length, so the tool speeds up the voice, slows the picture, or both. ### Dubbing with AI lip sync AI lip sync goes one step further and alters the speaker's mouth so it matches the new audio. Without it, a dubbed talking head looks slightly off, because the lips still form the original words. Lip sync only matters when a face is on screen and large enough to notice. Even YouTube describes its own version as a pilot that subtly matches lip movements to the translated audio [2]. ## The method: from transcript to translated video Whatever tool you end up using, the same sequence decides the quality. Skipping a step rarely saves time. It moves the error downstream, where it is copied into every language. 1. **Confirm the source language.** Set the original language correctly and clean up noisy audio if you can. YouTube warns that a wrongly set original language can produce mistranslated dubs [3]. 2. **Get an accurate transcript.** Every level starts here. Fix names, product terms and numbers before anything is translated. 3. **Build a short glossary.** List the terms that must stay fixed, such as brand names, job titles and technical words, with the agreed version in each target language. 4. **Translate, then condense.** Treat machine translation as a first draft. Trim lines to fit reading speed for subtitles, or the spoken length of each shot for dubbing. 5. **Have a fluent speaker review it.** One read by a native speaker catches tone, idioms and false friends that no tool flags. 6. **Produce the output.** Export the caption files, generate the dub, or run lip sync on the scenes where a face carries the message. 7. **Rebuild on-screen text.** Titles, slides and burned-in captions stay in the original language unless you replace them. 8. **Watch the result end to end.** Play it at normal speed, on a phone, with the sound on. Steps 2 to 5 are where a translation is won or lost. A tool can automate them in seconds, but it cannot know that your company calls its benefits portal by a nickname, or that a joke lands differently in Mexico City than in Madrid. Plan review time per language. Not per video. > 💡 **Write the glossary first:** Fix brand names, job titles and technical terms in every target language before translating. It is the simplest way to stop the same error repeating across versions. ## A worked example: one onboarding video, three languages Picture an HR manager with a six-minute onboarding video. One presenter speaks to camera for the opening and the closing minute. The middle is a slide walk-through with a voiceover. The company needs Spanish, French and Brazilian Portuguese versions before the next hiring wave. The transcript comes first, corrected for the company name, the benefit plan names and every acronym. A one-page glossary fixes how each term is said in all three languages. The translation is drafted by machine, then condensed so no subtitle line runs long. A bilingual colleague in each language reads the script before anything is voiced. Now the levels split by scene. **The slide section only needs a dubbed voice**, because nobody's mouth is visible. The opening and closing, where the presenter fills the frame, get lip sync so the face matches the new words. The slide text is rebuilt per language, with room left for longer lines. Each version also ships with its subtitle track, so viewers can read along or watch on mute. That is the whole job. The result is three localized videos with one review pass each. Most of the effort went into the transcript, the glossary and the human read, not into pressing a button. ## Common mistakes when you translate video Most failed translations fail for ordinary reasons. These are the ones that show up again and again: - **Word-for-word translation.** It produces subtitles that are too long to read and dubs that run past the scene. - **No human review.** Machine output sounds fluent even when it is wrong, which makes errors harder to spot. - **Forgetting on-screen text.** The voice speaks Spanish while the slide still says *Welcome aboard*. - **Ignoring timing.** A rushed dub sounds robotic; a stretched one drags. - **Cloning a voice without permission.** You need the right to use a person's voice and likeness, so get the speaker's consent before you clone or lip-sync them. - **Lip sync by default.** On wide shots, screen recordings or animation, it adds processing and risk with nothing to gain. Proper nouns, idioms and jargon are a known weak spot. YouTube says its automatic dubs can contain errors from mispronunciations, accents, dialects and background noise, and that proper nouns, idioms and jargon are hard to translate [3]. That is a fair description of AI dubbing in general. > ⚠ **Burned-in text stays untranslated:** Dubbing usually translates speech only. Titles, slides and captions baked into the picture have to be rebuilt in each language. ## When a video translator saves real time, and where AI lip sync stops Doing it by hand works for one video in one language. Scale changes that fast. A team with twenty training videos and five languages faces a hundred versions, each with its own captions, audio and review. That is where a video translator earns its place: it runs transcription, translation, voicing and timing as one job and keeps each language as a separate version you can check. Four situations push people toward a dedicated tool. **Many videos or many languages.** Several speakers who each need a distinct voice. A presenter whose own voice should carry over through cloning. And faces on camera, where AI lip sync is the only way to keep the picture believable. Limits remain. Automatic dubbing can skip videos with almost no speech, mostly music, or speech so fast that a dub would sound sped up; YouTube lists exactly those cases, plus videos longer than 120 minutes, as ineligible for its own feature. AI lip sync needs a clearly visible face, and it is the level where a small error looks worst. Specialist vocabulary still needs a person who knows the field. > 💡 **Lip-sync only on-camera scenes:** Use AI lip sync where a face fills the frame and plain dubbing everywhere else. Screen recordings and slides gain nothing from it. ## Handing the onboarding job to a dubbing tool Dedicated tools bundle the method into one flow, but they still depend on your transcript check and your review. For the HR manager above, whose presenter should still sound like the presenter in all three languages, Synthesia [dubs an uploaded video or a YouTube link with a clone of the speaker's voice](/articles/synthesia-video-translation). ### Synthesia — a presenter who keeps their voice AI Dubbing takes any uploaded video or a YouTube link, transcribes and translates the speech, and voices it with a clone of each original speaker; lip sync is an option in the dub. Videos made in Synthesia can also be translated with 1-Click Translation, which keeps the avatar and scenes. 1. Pick the source file or YouTube link 2. Choose the target language and whether to add lip sync 3. Check each language before publishing - Where it stops: Translates spoken audio only, not on-screen text or burned-in captions; on Basic, dubbed videos are watermarked and there is no MP4 download. - Plan: AI Dubbing is listed on all plans; 1-Click Translation is Enterprise - [Dubbing vs 1-Click Translation](https://professionalstoolkit.com/articles/synthesia-video-translation) · [Synthesia](https://professionalstoolkit.com/tools/synthesia) A team that wants the subtitle track as well as the dub, the way each onboarding version shipped with one, gets both from InVideo AI, which transcribes, translates and [voices the dub with text to speech or a consented voice clone](/articles/invideo-ai-voice-cloning). ### InVideo AI — dub, lip sync and subtitle files together InVideo AI says its Editor translates and dubs a finished cut with lip sync built in: the agent transcribes, translates, voices the script with text to speech or a consented clone, and re-syncs lips where a mouth shows. It also writes SRT or VTT caption files. - Where it stops: Text burned into the footage is not translated; voice cloning needs the speaker's explicit consent in a separate step. - [Voices and clones for dubbing](https://professionalstoolkit.com/articles/invideo-ai-voice-cloning) · [InVideo AI](https://professionalstoolkit.com/tools/invideo-ai) The third option covers less of the chain. Higgsfield re-voices finished videos and keeps lip sync in a separate studio, so the scene-by-scene split from the example becomes two passes you run yourself. ### Higgsfield — voice and lips, in two separate passes Higgsfield Audio's Translate re-voices a finished video in another language, and Lipsync Studio re-syncs the lips to the new audio with a lip-sync model you choose. - Where it stops: Voice cloning is for consented voices only. - Plan: Starter has one lip-sync model; the others start at Plus - [Higgsfield](https://professionalstoolkit.com/tools/higgsfield) If you would rather start free, [translate a video yourself with a transcript, a caption file and YouTube Studio](/articles/how-to-translate-a-video). If you already know you need a tool, our guide to choosing an [AI video translator for your situation](/articles/ai-video-translator) groups the options by who they fit. ## FAQ **Is there a free video translator?** If you publish on YouTube, automatic dubbing is enabled by default for eligible creators, and you can translate caption files yourself. Dedicated tools often start with a free plan; on Synthesia's Basic plan, for example, dubbed videos are watermarked and there is no MP4 download. **Can AI translate a video and keep my own voice?** Yes. Many AI dubbing tools clone the original speaker's voice and use it in the new language. You need the right to use that voice, and tools such as InVideo AI and Higgsfield only clone voices with the speaker's consent. **Does AI lip sync work on any video?** No. It needs a person whose mouth is clearly visible; Kling AI's Lip Sync, for example, needs a human character with the full face in view and does not work on animals. On wide shots and voiceover scenes, plain dubbing is enough. **How do I translate a YouTube video to English automatically?** YouTube's automatic dubbing creates translated audio tracks for eligible videos whose original language YouTube supports. In YouTube Studio, under Settings, Channel, Advanced settings, you can tick Manually review dubs before publishing; automatic dubs themselves cannot be edited. ## Matching the level to how your viewers watch Before any tool comes the method. Start from an accurate transcript and a short glossary, translate, condense, and have a fluent speaker read every language. Then choose the level by how people watch: subtitles for reading and mute viewing, a dubbed voice for listening, AI lip sync only where a face fills the frame. Rebuild on-screen text separately and get consent before cloning anyone's voice. Done that way, a single video can serve several audiences without looking like a translation. ## Sources [1] Netflix Partner Help Center — https://partnerhelp.netflixstudios.com/hc/en-us/articles/217350977-English-USA-Timed-Text-Style-Guide (Nov 2026) [2] YouTube Official Blog — https://blog.youtube/news-and-events/youtube-auto-dubbing-expressive-speech/ (Nov 2026) [3] YouTube Help — https://support.google.com/youtube/answer/15569972 (Nov 2026)